• The Journal of the Acoustical Society of Korea
• /
• v.42 no.6
• /
• pp.544-551
• /
• 2023

Speech enhancement used to improve the perceptual quality and intelligibility of noise speech has been studied as a method using a complex-valued spectrum that can improve both magnitude and phase in a method using a magnitude spectrum. In this paper, a study was conducted on how to apply attention mechanism to complex-valued spectrum-based speech enhancement systems to further improve the intelligibility and quality of noise speech. The attention is performed based on additive attention and allows the attention weight to be calculated in consideration of the complex-valued spectrum. In addition, the global average pooling was used to consider the importance of the feature map. Complex-valued spectrum-based speech enhancement was performed based on the Deep Complex U-Net (DCUNET) model, and additive attention was conducted based on the proposed method in the Attention U-Net model. The results of the experiments on noise speech in a living room environment showed that the proposed method is improved performance over the baseline model according to evaluation metrics such as Source to Distortion Ratio (SDR), Perceptual Evaluation of Speech Quality (PESQ), and Short Time Object Intelligence (STOI), and consistently improved performance across various background noise environments and low Signal-to-Noise Ratio (SNR) conditions. Through this, the proposed speech enhancement system demonstrated its effectiveness in improving the intelligibility and quality of noisy speech.

• Journal of Intelligence and Information Systems
• /
• v.29 no.4
• /
• pp.165-184
• /
• 2023

Recently, Deep learning analysis of unstructured text data using language models, such as Google's BERT and OpenAI's GPT has shown remarkable results in various applications. Most language models are used to learn generalized linguistic information from pre-training data and then update their weights for downstream tasks through a fine-tuning process. However, some concerns have been raised that privacy may be violated in the process of using these language models, i.e., data privacy may be violated when data owner provides large amounts of data to the model owner to perform fine-tuning of the language model. Conversely, when the model owner discloses the entire model to the data owner, the structure and weights of the model are disclosed, which may violate the privacy of the model. The concept of offsite tuning has been recently proposed to perform fine-tuning of language models while protecting privacy in such situations. But the study has a limitation that it does not provide a concrete way to apply the proposed methodology to text classification models. In this study, we propose a concrete method to apply offsite tuning with an additional classifier to protect the privacy of the model and data when performing multi-classification fine-tuning on Korean documents. To evaluate the performance of the proposed methodology, we conducted experiments on about 200,000 Korean documents from five major fields, ICT, electrical, electronic, mechanical, and medical, provided by AIHub, and found that the proposed plug-in model outperforms the zero-shot model and the offsite model in terms of classification accuracy.

• Korean Journal of Construction Engineering and Management
• /
• v.25 no.3
• /
• pp.3-16
• /
• 2024

The government announced the Housing Welfare Roadmap (November 2017), to expand the supply of public rental housing by reconstructing aged long-term public rental complexes. Also, remodeling projects for complexes with low business feasibility of reconstruction projects are recognized as an alternative to supplying public rental housing in urban area. This study analyzed influence factors by dividing them into project feasibility, architectural plan, urban & residential environment plan, and legal system groups in order to establish a plan for long-term public rental housing remodeling project. Futhermore, this work conducted the principal component analysis to get the principal component factors among the influence factors of each group, and the weight analysis to calculate weighting of them. In addition, major influence factors were derived by calculating the relative importance score (RIS) of each factor. Lastly this paper validated the major influence factors and applicability of the procedure to select 3 complexes that can be reviewed for remolding project among 33 long-term public rental housing complexes located in Seoul. The results of this study are expected to be useful when establishing a remodeling project plan for long-term public rental housing.

• Korean Journal of Soil Science and Fertilizer
• /
• v.45 no.6
• /
• pp.1143-1152
• /
• 2012

We established a top-down methodology to estimate carbon footprint as national mean value (reference) with the statistical data on agri-livestock incomes in 2007. We also established LCI (life cycle inventory) DB by a bottom-up methodology with the data obtained from interview with farmers from 4 large-scale farms at Gunsan, Jeollabuk-do province to estimate carbon footprint in 2011. This study was carried out to compare top-down methodology and bottom-up methodology in performing LCA (life cycle assessment) to analyze the difference in GHGs (greenhouse gases) emission and carbon footprint under conventional rice cultivation system. Results of LCI analysis showed that most of $CO_2$ was emitted during fertilizer production and rice cultivation, whereas $CH_4$ and $N_2O$ were mostly emitted during rice cultivation. The carbon footprints on conventional rice production system were 2.39E+00 kg $CO_2$-eq. $kg^{-1}$ by top-down methodology, whereas 1.04E+00 kg $CO_2$-eq. $kg^{-1}$ by bottom-up methodology. The amount of agro-materials input during the entire rice cultivation for the two methodologies was similar. The amount of agro-materials input for the bottom-up methodology was sometimes greater than that for top-down methodology. While carbon footprint by the bottom-up methodology was smaller than that by the top-down methodology due to higher yield per cropping season by the bottom-up methodology. Under the conventional rice production system, fertilizer production showed the highest contribution to the environmental impacts on most categories except GWP (global warming potential) category. Rice cultivation was the highest contribution to the environmental impacts on GWP category under the conventional rice production system. The main factors of carbon footprints under the conventional rice production system were $CH_4$ emission from rice paddy field, the amount of fertilizer input and rice yield. Results of this study will be used for establishing baseline data for estimating carbon footprint from 'low carbon certification pilot project' as well as for developing farming methods of reducing $CO_2$ emission from rice paddy fields.

• Korean Journal of Soil Science and Fertilizer
• /
• v.37 no.4
• /
• pp.199-206
• /
• 2004

Factors of universal soil loss equation, USLE, and its revised version, RUSLE for Korean soils were reevaluated to estimate the national scale of soil loss based on digital soil maps. Rainfall erosivity factor, R, of 158 locations of cities and counties were spacially interpolated by the inverse distance weight method. Soil erodibility factor, K, of 1321 soil phases of 390 soil series were calculated using the data of soil survey and agri-environmental quality monitoring. Topographic factor, LS, was estimated using soil map of 1:25,000 scale with soil phase and land use type. Cover management factor, C, of major crops and support practice factor, P, were summarized by analyzing the data of lysimeter and field experiments for 27 years (1975-2001) in the National Institute of Agricultural Science and Technology. R factor varied between 2322 and 6408 MJ mm $ha^{-1}$ $yr^{-1}$ $hr^{-1}$ and the average value was 4276 MJ mm $ha^{-1}$ $yr^{-1}$ $hr^{-1}$. The average K value was evaluated as 0.027 MT hr $MJ^{-1}$ $mm^{-1}$. The highest K factor was found in paddy rice fields, 0.034 MT hr $MJ^{-1}$ $mm^{-1}$, and K factors in upland fields, grassland, and forest were 0.026, 0.019, and 0.020 MT hr $MJ^{-1}$ $mm^{-1}$, respectively. C factors of upland crops ranged from 0.06 to 0.45 and that of grassland was 0.003. P factor varied between 0.01 and 0.85.

• Journal of the Korean Society of Radiology
• /
• v.14 no.4
• /
• pp.467-475
• /
• 2020

The purpose of this study is to assess doses to ¹⁸F-FDG, a radioactive drug, during PET examinations, to alleviate anxiety about radiation in patients and carers, to minimize the indiscriminate examination progress caused by medical institution personnel and space clearance problems, and health examination. The dose assessment was measured using a thermo-fluorescent dosimeter (TLD) and an electronic personal dosimeter (EPD) at the location of the cervical (hypothyroid), thorax (heart), and lower abdomen (breeding line) which are the three highest tissue areas of the radiation tissue weighting. In addition, spatial dose rates and radioactivity in urine were measured using GM counters and ion boxes. The results are as follows: First, the personal dosimeter TLD was measured 0.0425±0.0277 mSv in the cervical region, 0.0440±0.0386 mSv in the thorax and 0.0485±0.0436 mSv in the lower abdomen, with little difference in the heart dose depending on radiation sensitivity. The EPD was measured at 0.942±0.141 mSv/h immediately after the cervical position, and 0.192±0.031 mSv/h after 120 minutes. Immediately after the thorax position, 0.516±0.085 mSv/h, 120 minutes later 0.128±0.040 mSv/h. Immediately after the lower abdomen position, 0.468±0.091 mSv/h, and after 120 minutes 0.105±0.021 mSv/h were measured. The spatial dose rate at the GM counter was measured immediately at 0.041±0.005 mSv/h, 120 minutes later at 0.014±0.002 mSv/h. The radioactivity in urine using ion chamber was measured at 0.113±0.24 MBq/cc after 60 minutes and 0.063±0.13 MBq/cc after 120 minutes. As a result, ¹⁸F-FDG should be administered, dose re-evaluated two hours after the PET test is completed, and caregivers should be avoided. In addition, it is deemed necessary to provide patients and carers with sufficient explanations and expected values of exposure dose to avoid reckless testing. It is hoped that the data tested in this study will help patients and families relieve anxiety about radiation, and that the radiation workers' exposure management system and institutional improvements will contribute to the development of medical radiation.

• Korean Journal of Soil Science and Fertilizer
• /
• v.31 no.1
• /
• pp.16-24
• /
• 1998

To determine the optimum application of fertilizers for the cultivation of Chinese cabbage in plastic film house, twenty soils which contain different salts contents were taken from 4 different area of plastic film house cultivation, Youngdong. Boeun county, Cheongweon county, and Cheongju city. The dry weight and the amount of N. P, and K uptakes of Chinese cabbage in the plot of no fertilization were considered as the factors representing the fertility of the soil. And a difference of dry weight and the amounts of N, P, and K uptakes of plants between the plot of fertilization and no fertilization were considered as the factors representing the total effect of fertilizer and fertilizer N, P, and K effects. respectively. These factors of soil fertility and fertilizer effects were estimated by correlation and regression with soil tests in order to find the critical levels and recommended method for optimum fertilization of Chinese cabbage. Chinese cabbage transplanted in two soils, having the electrical conductivity of 9.3 and 15.2 dS/m, were not able to root due to the salts toxicity. The content of inorganic N, the electrical conductivity, and CEC were founded to have significant correlation with the factors of both the soil fertility and fertilizer effects for the cultivation of Chinese cabbage. To determine the weighting degree for the productivity and the fertilizer effects, the standardized partial regression coefficient was analyzed by regression among the factors of fertility, the fertilizer effects, and the soil tests. The coefficient for inorganic N($NH_4-N$ and $NO_3-N$) was obtained as the absolute value of 756-1871 and this value was extremely higher than those of other soil tests which was 0.07-4.11. These results suggested that the content of inorganic N is the best tests for the estimation of the productivity and the fertilizer effects for the cultivation of Chinese cabbage in plastic film house. The critical level of inorganic N($NH_4-N+NO_3-N$) estimated by Cate-Nelson split method for maximum productivity and zero point of fertilizer effect was 220 mg/kg for all the factors of estimation. These results suggested that no application of fertilizer N. P, and K is required at the critical level of inorganic N of soil. Consequently the optimum application of fertilizer N, P, and K for the cultivation of Chinese cabbage in plastic film house was possible to determine by the critical level of inorganic N of soil. The critical level of electrical conductivity was estimated as 2.8 dS/m by the same method.

• Korean Journal of Remote Sensing
• /
• v.39 no.3
• /
• pp.283-295
• /
• 2023

With the development of sensor and satellite technology, numerous high-resolution and multi-spectral satellite images have been available. Due to their wavelength-dependent reflection, transmission, and scattering characteristics, multi-spectral satellite images can provide complementary information for earth observation. In particular, the short-wave infrared (SWIR) band can penetrate certain types of atmospheric aerosols from the benefit of the reduced Rayleigh scattering effect, which allows for a clearer view and more detailed information to be captured from hazed surfaces compared to the visible band. In this study, we proposed a multi-resolution transform-based image fusion method to combine visible and SWIR satellite images. The purpose of the fusion method is to generate a single integrated image that incorporates complementary information such as detailed background information from the visible band and land cover information in the haze region from the SWIR band. For this purpose, this study applied the Laplacian pyramid-based multi-resolution transform method, which is a representative image decomposition approach for image fusion. Additionally, we modified the multiresolution fusion method by combining a haze-guided weight map based on the prior knowledge that SWIR bands contain more information in pixels from the haze region. The proposed method was validated using very high-resolution satellite images from Worldview-3, containing multi-spectral visible and SWIR bands. The experimental data including hazed areas with limited visibility caused by smoke from wildfires was utilized to validate the penetration properties of the proposed fusion method. Both quantitative and visual evaluations were conducted using image quality assessment indices. The results showed that the bright features from the SWIR bands in the hazed areas were successfully fused into the integrated feature maps without any loss of detailed information from the visible bands.

• Journal of Intelligence and Information Systems
• /
• v.23 no.2
• /
• pp.1-17
• /
• 2017

The deep learning framework is software designed to help develop deep learning models. Some of its important functions include "automatic differentiation" and "utilization of GPU". The list of popular deep learning framework includes Caffe (BVLC) and Theano (University of Montreal). And recently, Microsoft's deep learning framework, Microsoft Cognitive Toolkit, was released as open-source license, following Google's Tensorflow a year earlier. The early deep learning frameworks have been developed mainly for research at universities. Beginning with the inception of Tensorflow, however, it seems that companies such as Microsoft and Facebook have started to join the competition of framework development. Given the trend, Google and other companies are expected to continue investing in the deep learning framework to bring forward the initiative in the artificial intelligence business. From this point of view, we think it is a good time to compare some of deep learning frameworks. So we compare three deep learning frameworks which can be used as a Python library. Those are Google's Tensorflow, Microsoft's CNTK, and Theano which is sort of a predecessor of the preceding two. The most common and important function of deep learning frameworks is the ability to perform automatic differentiation. Basically all the mathematical expressions of deep learning models can be represented as computational graphs, which consist of nodes and edges. Partial derivatives on each edge of a computational graph can then be obtained. With the partial derivatives, we can let software compute differentiation of any node with respect to any variable by utilizing chain rule of Calculus. First of all, the convenience of coding is in the order of CNTK, Tensorflow, and Theano. The criterion is simply based on the lengths of the codes and the learning curve and the ease of coding are not the main concern. According to the criteria, Theano was the most difficult to implement with, and CNTK and Tensorflow were somewhat easier. With Tensorflow, we need to define weight variables and biases explicitly. The reason that CNTK and Tensorflow are easier to implement with is that those frameworks provide us with more abstraction than Theano. We, however, need to mention that low-level coding is not always bad. It gives us flexibility of coding. With the low-level coding such as in Theano, we can implement and test any new deep learning models or any new search methods that we can think of. The assessment of the execution speed of each framework is that there is not meaningful difference. According to the experiment, execution speeds of Theano and Tensorflow are very similar, although the experiment was limited to a CNN model. In the case of CNTK, the experimental environment was not maintained as the same. The code written in CNTK has to be run in PC environment without GPU where codes execute as much as 50 times slower than with GPU. But we concluded that the difference of execution speed was within the range of variation caused by the different hardware setup. In this study, we compared three types of deep learning framework: Theano, Tensorflow, and CNTK. According to Wikipedia, there are 12 available deep learning frameworks. And 15 different attributes differentiate each framework. Some of the important attributes would include interface language (Python, C ++, Java, etc.) and the availability of libraries on various deep learning models such as CNN, RNN, DBN, and etc. And if a user implements a large scale deep learning model, it will also be important to support multiple GPU or multiple servers. Also, if you are learning the deep learning model, it would also be important if there are enough examples and references.

• Journal of Animal Science and Technology
• /
• v.55 no.1
• /
• pp.7-11
• /
• 2013

The purpose of this study was to estimate (co) variance components of three milk production traits for genetic evaluation using a multiple lactation model. Each of the first five lactations was treated as different traits. For the parameter estimation study, a data set was set up including lactations from cows calved from 2001 to 2009. The total number of raw lactation records in first to fifth parities reached 1,416,589. At least 10 cows were required for each contemporary group, herd-year-season effect. Sires with fewer than 10 daughters were discarded. Lactations with 305d milk yield exceeding 15,000 kg were removed. In total, 1,456 sires of cows were remained after all the selection steps. A complete pedigree consisting of 292,382 records was used for the study. A sire model containing herd-year-season, caving age, and sire additive genetic effects was applied to the selected lactation data and pedigree for estimating (co) variance components via VCE. Heritabilities and genetic or residual correlations were then derived from the (co) variance estimates using R package. Genetic correlations between lactations ranged from 0.76 to 0.98 for milk yield, 0.79~1.00 for fat yield, 0.75~1.00 for protein yield. On individual lactation basis, relatively low heritability values were obtained 0.14~0.23, 0.13~0.20 and 0.14~0.19 for milk, fat, and protein yields, respectively. For the combined lactation heritability values were 0.29, 0.28, and 0.26 for milk, fat, and protein yields. The estimated parameters will be used in national genetic evaluations for production traits.